Individual Differences in Multisensory Attention Skills in Children with Autism Spectrum Disorder Predict Language and Symptom Severity: Evidence from the Multisensory Attention Assessment Protocol (MAAP)

We use “multisensory” as a general term to refer to stimulation impacting more than one sensory system (e.g., auditory, visual, proprioceptive, etc.). It serves as a name for our protocol (MAAP) and for the collection of the three attention skills it measures (“multisensory attention skills” (MASks): attention maintenance, shifting/disengaging, intersensory matching). In contrast, the term “intersensory” is more specific and is used here to refer to just one of these skills—intersensory matching. Intersensory matching is the detection of information that is common across auditory and visual stimulation such as synchrony, rhythm, tempo, or intensity patterns. We also refer to intersensory processing as the activity of perceiving, integrating, and further processing this information.

Assuming a β of .80 and a two-tailed p value of .05, a sample size of n = 21 per group (ASD, TD; total N = 42) has sufficient power to detect effect sizes of 1) d = .88 for between-group differences, 2) d = .76 for within-group differences (assuming a .30 correlation of scores between groups), and 3) r = .54 for correlation analyses.

Preliminary analyses revealed little evidence of effects of presentation order on the three MASks. Thus, we chose not to include presentation order as a factor in our main analyses. For details, see Supplemental Material, p. 4.

Given that social neutral (n = 8) and social positive (n = 8) trials were collapsed into a single condition, there were a greater number of social (n = 16 trials), than nonsocial trials (n = 8) overall. It is thus possible that estimates of MASks for social events may have been more stable, and less impacted by outliers, than MASks for nonsocial events. However, analyses revealed little difference in individual variability in MASks for social or nonsocial events. We calculated the coefficient of variation (CV; a scale independent index of variability) for each MASk. On average, for TD children, the CVs for social and nonsocial events were equivalent (social: 32.41, range: 14.29–82.08; nonsocial: 36.44, range: 13.46–108.49) suggesting no difference in MAAP performance on social and nonsocial trials. In contrast, for children with ASD, the average CV for social events was larger than for nonsocial events (social: 43.93, range: 21.57–71.82; nonsocial: 36.39, range: 12.00–97.18).

Prior to conducting our main analyses, we conducted preliminary analyses to determine if home language, sex, and ethnicity would be important covariates to include in our analyses. Also, given that MA-matched TD children had higher MSEL Verbal Age Equivalence scores than children with ASD, we also assessed whether Verbal scores should be included as covariates in our analyses. Results of these preliminary analyses indicate that the inclusion of home language, sex, ethnicity, and Verbal Age Equivalence Scores did not qualify our results, and thus they were not included as covariates in the main analyses (for details, see Supplemental Material, pp. 4–5).

Significantly slower shift speeds on high than low competition trials were evident for TD children when both MA- and CA-matched TD children were analyzed as a single TD group (n = 33), F(1, 32) = 8.13, p = .01. Thus, the lack of a significant difference for MA-matched TD children was likely due to decreased statistical power due to a small sample size.

Our prior research using the MAAP demonstrates significant intersensory matching of nonsocial events in both 2–5-year-old TD children, p = .05, and 12-month-old TD infants, p = .01 (Bahrick et al., 2018a, 2018b; pp. 2213, 2218).

We created a family of four for each language outcome (Receptive, Expressive Language) given there are four possible correlations in our 2 event (social, nonsocial) × 2 competition (high, low) design (see Table 5), Thus, the correlation with the smallest p value is compared against a critical value of p < .0125 (.05/4). If the correlation with the smallest p value is less than .0125, it is declared significant. Then, the correlation with the next smallest p value is compared against a critical value of p < .0167 (.05/3), and so on.

Alternative models were tested but ultimately rejected. For children with ASD, a model including Shift/Disengage as a separate predictor of Language, in addition to Maintenance and Intersensory, failed to predict variance in language over and above that predicted by just Maintenance and Intersensory, R² change = .06, p = .17. Also, for children with ASD we conducted the model depicted in Fig. 3A but using nonsocial trials. Unlike the model with social events, this nonsocial model only predicted a nonsignificant 6% of the variance on Language, p = .59, and neither Maintenance nor Intersensory Matching of nonsocial events were significant predictors of Language, ps > .32. Thus, none of the alternative models using attention to social or nonsocial events improved our ability to predict language outcomes in children with ASD.

For TD children, we also tested the model depicted in Fig. 3B but using nonsocial events. Results indicated no significant pathways among MASks for nonsocial events. Further, unlike our model with social events, intersensory matching for nonsocial events was not a significant predictor of language, p = .57. Thus, the alternative model using attention to nonsocial events did not improve our ability to predict language outcomes in TD children.